C6100 strange issue: node different behavior in different enclosure

korban

Member
Feb 12, 2013
33
1
8
Belgium
www.befr.ebay.be
Hello,

I'm back with my strange things :p

I always have 2 C6100 chassis, each with 4 nodes.

The L5520 have a cool function, it clock down to save power or it clock up (intel turbo mode) on the high demand.
The CPU play with his multiplier to change the CPU core frequency on the fly. Until now nothing new :)

In my C6100 the L5520 has a range of 12 to 17 multiplier. Turbo mode goes to 19.

When I use CPU-Z to check the beast, I'm happy to see the frequency varying from 1600mhz with a 12 to 2500 with a 19.

When I received the chassis I have benchmarked and stressed each node... Ans was soooo happy.

I had a very small fan issue and try to solve with a full firmware update ( mbc + fbc + bios). All was ok excepted in the chassi2, the fbc is a pic16 and I was not able to upgrade with the official c6100 firmware . I deal with this and pass my way.

Now the best part for those who have read to here ;)

From a week, I have noticed a perf reduction on the node on wich I'm working. I benchmark back the node and ... The multiplier stay blocked @ 12 then @ 1600 MHz .

I have checked the bios, the os, the ram, deferent nodes... I boot always the same os on the same ssd.

Now the tricky part. If I take one node from the chassi1 and plug into the chassi2... The frequency vary well with the load. If I put back the node from the chassi2 to chassi1... The issue is back.

The only delta on both chassis is the fcb, the I rolled back the firmware of the chassi1 from 1.20 to 1.17 (was the original version for me)

Thanks for your help I'm not able to understand why a same node has a different CPU behavior in a different enclosure :(

Just for fun I have screencasted the thing

Chassi2: the good one
https://dl.dropboxusercontent.com/u/4395168/1.flv
https://dl.dropboxusercontent.com/u/4395168/1.gif

Chassi1: the bad one
https://dl.dropboxusercontent.com/u/4395168/2.flv
https://dl.dropboxusercontent.com/u/4395168/2.gif

Always same node, same ssd, same os, the only difference is the enclosure where the node is plugged
 
Last edited:

Patrick

Administrator
Staff member
Dec 21, 2010
11,970
4,931
113
Haven't seen this one yet :-/ Looks super strange. At first was thinking power management.
 

korban

Member
Feb 12, 2013
33
1
8
Belgium
www.befr.ebay.be
Haven't seen this one yet :-/ Looks super strange. At first was thinking power management.
I have the same idea, but I boot the same OS and the same node. Only the chassi is changing.

I have also sweped the PSU and put 2 PSU in the "bad" node = no changes.

Have you a clue where I can look in the chassi?
 

PigLover

Moderator
Jan 26, 2011
2,969
1,280
113
The node's power management is handled by the Fan Control Board. From your description above it sounds like you have two different FCB models. They may be behaving differently - and in fact one of them may be defective.

Give this a try (this is based on a guess only). In the BIOS Server menu on one of your nodes under "power management" change it from the default value ("node management") to "Operating system Control/EIST". This should disable the node's power management interface and allow EIST to be controlled completely by the OS.

Plug this node into each chassis and see if you still have the multiplier limits or if they are behaving normally. If they work you can set all of the nodes this way or - better - replace the faulty FCB and go back to node management.

If the behavior doesn't change then I guess that wasn't the problem...